backtop

Print E-mail del.icio.us 80 comment(s) - last by Hieyeck.. on Sep 10 at 2:22 AM

Turbo Memory's successor, Braidwood is set to debut in 2010 on Westmere motherboards. Intel has put significant work into improving its drivers to offer better performance.  (Source: CNET)

Some say Braidwood may negate much of the performance advantages of SSD drives. Some analysts are saying it could seriously damage the fledgling SSD market, whose growth is primarily being driven by the performance edge it offers.  (Source: OCZ)
Is Intel's new motherboard flash poised to damage SSD demand?

Turbo memory is returning in 2010, says chipmaker Intel.  Intel's original Turbo Memory, which first debuted in 2006, was less than well received.  While it shipped millions of units by Intel's own estimates, reviews of the product were lukewarm.  Armed with new drivers and a new plan, this time around Intel believes it has what it takes to achieve a much greater success.

Intel's upcoming Turbo Memory successor, Braidwood, will consist of NAND flash module residing on "5 Series" motherboards (used with the upcoming Westmere 32 nm processors) and serve as a cache for all reads and writes.  Capacities will be approximately 4GB to 16GB, and the cost increase will be approximately $10 to $20 per system, according to analyst Jim Handy, who authored a recent report on Braidwood.  The technology is set to launch in the first quarter of 2010, though it may be delayed.

Some former skeptics of Turbo Memory have become Braidwood believers.  Some are even going as far as to say that it could send the burgeoning SSD market reeling.  Mr. Handy is among those convinced that the new product will trouble solid state drive markers.  He points out that the new cache uses SLC (single level cell) NAND, which is approximately a quarter of the cost of the DRAM traditionally used in caches.  Meanwhile, it provides better performance than most solid state drives, which use the cheaper, but lower performance MLC (multi-level cell) NAND.

He states, "The move to NAND in PCs will boost the NAND market, soften the SSD and DRAM markets and pose problems for those NAND makers who are not poised to produce ONFi (open NAND flash interface) NAND flash."

Traditionally, performance has been the strongest selling point of SSDs.  Other benefits include lower power consumption and increased reliability over hard disc drives.

If the SSD market suffers, Intel could be hurting itself.  Intel currently makes two relatively well selling drives -- the X25-M and the X25-E.  The company, however, disagrees with Mr. Handy's analysis.

Intel responded to the analyst's remarks, stating, "It's not just the performance, but also the added reliability...[SSDs] can help facilitate versus a hard drive. We see a long life ahead for SSDs, and won't stop inventing a variety of other technologies that make computers faster and/or more energy efficient."

However, Mr. Handy counters, "If you really get down to what makes consumers buy SSD, the reliability issue is not something they often cite as reason [for] spending extra money on an SSD."

According to Mr. Handy current SSD makers -- Toshiba, Samsung, Hynix, Micron -- as well as DRAM suppliers will be most effected.  He believes that if SSDs no longer offer significantly superior performance, few will buy them for their improved reliability and lower power consumption, when hard drive power consumption is already low in comparison to other system components and most drives are already relatively reliable.

He concludes, "Intel has got a very good [SSD] product. But, they view additional layers of NAND technology in PCs as inevitable. They don't think SSDs are likely to take over 100% of the PC market, but they do think Braidwood could find itself in 100% of PCs."

Comments     Threshold


This article is over a month old, voting and posting comments is disabled
Bottleneck
By Mitch101 on 9/4/2009 10:19:20 AM , Rating: 4
I am so glad to see that the one major bottleneck in most PC's for the last X years is finally getting some major progress and speed improvements. Yea platter densities improved and rpms were cranked up for home machines and were finally seeing the benefits of SATA over EIDE but man does SSD rock and provide a big jump in performance. Now if it was just a bit more affordable and my core OS install smaller but Im salivating that its sooner than later.




RE: Bottleneck
By tjr508 on 9/4/2009 10:29:18 AM , Rating: 2
$100-110 for an 30G Vertex isn't affordable? 3-4 Wal-Mart meals instead of dining out and your PC can feel twice as fast...


RE: Bottleneck
By therealnickdanger on 9/4/2009 10:45:23 AM , Rating: 2
Haha, so true!

However, be it Vista or W7, you'll devour half the capacity of that drive just loading the OS. Load Office, CS4, and one game and you'll be full. The best option is to go for the 60/64GB versions of the Indilix drives: the cheapest I've seen is the Super Talent Ultradrive ME for $190 and the Corsair version is about $180 after rebates...

Or you could pay a little more and get 2x30GB Vertex for RAID...


RE: Bottleneck
By Trippytiger on 9/4/2009 10:58:24 AM , Rating: 2
I picked up a 64GB Ultradrive for $150 CAD a few weeks ago from Newegg. That was a great price. At the current, non-sale prices, though, I think the Intel drives are the way to go.


RE: Bottleneck
By nemitech on 9/4/2009 11:05:35 AM , Rating: 2
quote:
The best option is to go for the 60/64GB versions of the Indilix drives: the cheapest I've seen is the Super Talent Ultradrive ME for $190 and the Corsair version is about $180 after rebates... Or you could pay a little more and get 2x30GB Vertex for RAID...


I disagree, the best option is to go for the 120Gb drives as they have greater read/write speeds than the 60Gb and 30 Gb versions:
For $234 after rebates I got a 120Gb OCZ Agility (indilinx controller).


RE: Bottleneck
By Mitch101 on 9/4/2009 11:23:55 AM , Rating: 2
The cheapest I can find the 120Gb OCZ Agility is around $300.00. You got a nice deal.

Thats close to what I am looking for $200 and around 120 gig but would like to get two 64gig and raid-0 them for max performance but could settle on a single 120gig when around $200.00

I run Windows 7 ultimate and with everything installed 64gig won't cut it.


RE: Bottleneck
By PrezWeezy on 9/8/2009 8:41:11 PM , Rating: 2
Just remember you'll need a good controller to actually take advantage of that performance boost.


RE: Bottleneck
By teldar on 9/4/2009 7:33:39 PM , Rating: 3
So only put your OS on it and the two or three programs you use the most. Install EVERYTHING else on a different drive. One of the green series or something with just decent performance. Then you'll have a super fast OS and decent load times for everything as the computer is accessing multiple drives for multiple files instead of accessing one drive for all the files.


RE: Bottleneck
By jonmcc33 on 9/5/2009 11:13:41 PM , Rating: 3
$190 and that's 2TB of data for me. Considering my extreme downloading habits I'll take the capacity over the speed.


RE: Bottleneck
By Hieyeck on 9/10/2009 2:22:13 AM , Rating: 2
Or get both...?

Dropped an SSD onto my lappy, and it runs faster than my desktop. And with a the fileserver I just slapped together, I'm golden.


RE: Bottleneck
By inighthawki on 9/4/2009 10:46:02 AM , Rating: 3
Affordable? Yeah these days they are getting there. A bargain, however...No. You are under the assumption that people must dine out often. In the case of my family, we would be lucky to dine out every other week, and that still wouldn't be anything fancy, that's for sure. Believe it or not there are quite a few people who don't have $100+ that can go for something as trivial as a faster hard drive. Sometimes it's not as easy as just cutting back a few things if you don't really have much to cut back on.


RE: Bottleneck
By Jackattak on 9/4/2009 11:28:23 AM , Rating: 3
I would hardly call the performance advantages of SSD's "trivial". SSD upgrades are, simply put, the single most cost-effective performance upgrade you can perform at this time.

I do agree with you on the eating out portion of your comment, though. In these days of ultimate economic recession, people are eating out less and less and it is a bit more difficult to save $100 than it used to be, that's for sure.

Just look at all the chain restaurant closures for evidence.


RE: Bottleneck
By Yawgm0th on 9/4/2009 2:39:48 PM , Rating: 1
quote:
I would hardly call the performance advantages of SSD's "trivial". SSD upgrades are, simply put, the single most cost-effective performance upgrade you can perform at this time.
RAID is still pretty darn cost effective. I can put four hard drives in a RAID10 or 0+1 array for less than the cost of a much smaller flash drive. They will also have redundancy and better throughout. That's still slower than a good SSD for "regular usage" tasks, but more cost-effective.


RE: Bottleneck
By FaaR on 9/7/2009 12:22:50 PM , Rating: 2
Better throughput only for linear reads and writes (and only marginally); the least important measure of the factors that determine I/O speed in a modern PC, be it desktop, workstation or server.

Linear read/writes are almost entirely without consequence when it comes to system I/O performance; when it takes a factor of 100s or 1000s of times longer to seek to a set of sectors than it takes to read or write said sectors it's pretty obvious RAID 0 is not going to make much of a difference in hitching up your PC's speed/responsiveness.

Repeat seek slowness by tens of thousands of times every time you boot up your PC and you see the lost time mount up hugely. I start up about 10-12 apps in my systray (some quite large and bloated, like MSN Messenger, Skype and Sony Ericsson PC Suite), and many undoubtedly more hidden as services and other background tasks.

If I were to launch a web browser or a game straight after logging in using a harddrive as the storage device the poor HDD would thrash its tiny little heads off for about 20-30 seconds before my app becomes somewhat responsive. With an Intel SLC SDD powering my main rig, the program or game is up and responsive in mere seconds right after logging in.

Any amount of RAID 0+1 drives aren't going to be able to compete with that level of performance...


RE: Bottleneck
By therealnickdanger on 9/4/2009 2:48:23 PM , Rating: 2
It's possible that the comment about cutting back on eating out simply doesn't apply to either you. Between my schedule and my old lady, we're lucky to eat at home even once per week. So if I could cut back on eating out, I could probably get a 4TB RAID of these things going! LOL

I certainly agree with your comment about the triviality of SSDs. You simply can not spend $200 and get a better overall performance upgrade right now than with an SSD. Even back when I bought my JMicron SSD (64GB G.Skill), it beat the living sh*t out of my 7200RPM HDD. The difference was, for lack of a better word, amazing. Once prices stabilize a little more on the Indilix and Intel drives, I'm going big-time.


RE: Bottleneck
By Jackattak on 9/4/2009 3:43:38 PM , Rating: 2
Oh believe me I used to eat out constantly, but it was more due to not feeling like cooking dinner (i.e. a totally asinine waste of money that needed to be curbed, anyhow).

Once my wife and I started cooking at home (when she lost her job like every other architect in Portland) we stymied that behavior and saved...literally...hundreds of dollars per month.

Still can't afford an SSD though. :)


RE: Bottleneck
By EricMartello on 9/4/2009 4:18:51 PM , Rating: 1
The performance of an SSD is hardly trivial if you are using it for business purposes where time is money. The dramatic performance boost to productivity can easily offset the higher cost of the drive itself.


RE: Bottleneck
By inighthawki on 9/4/2009 7:35:44 PM , Rating: 2
I apologize since I didn't make it entirely clear. I made some assumptions based on the OP that cutting down on food spending meant that the saved money would go towards a pleasure more so than a necessity or much needed improvement like in a business purpose. When I said trivial I also may have been over-exaggerating my point, since I understand the enormous speed benefits and that there are plenty of reasons why you would want an SSD over a normal hard drive. In the end I think I was more or less pointing out that based on the OP, it sounded like something unnecessary to be cutting food costs to buy.


RE: Bottleneck
By Nyu on 9/4/2009 2:14:18 PM , Rating: 2
make that 2x in Europe


RE: Bottleneck
By Yawgm0th on 9/4/2009 2:36:44 PM , Rating: 2
quote:
$100-110 for an 30G Vertex isn't affordable? 3-4 Wal-Mart meals instead of dining out and your PC can feel twice as fast...

Until you put something other than the operating system on it. Being faster doesn't make up for being 98% smaller.

...

I don't think that last sentence came out right...


RE: Bottleneck
By therealnickdanger on 9/4/2009 3:10:02 PM , Rating: 2
Depends on how she like it, I guess... ;-)

I thought I was crazy when I bought two 36GB Raptors for my RAID-0 about 4-5 years ago. I spent $500 on those drives (combined) and they were the hottest drives in town. That same setup today would be absolutely trounced by a single X-25M 80GB for half the price! (well, more than a third less, anyway)

Even then, like now, I keep all my media and personal files on a RAID server, so I don't need a lot of room on an SSD. For my own needs, a 30GB wouldn't cut it, but a 64GB would be too much. When I'm working with any files, they are stored locally (RAM and working directory), but then moved onto the server when complete.

One day, in the not-too-distant-future, multi-TB SSDs will be as affordable as conventional HDDs are today, so then people like you who store everything on one drive can be happy too. :)


RE: Bottleneck
By Oregonian2 on 9/4/2009 4:07:55 PM , Rating: 2
I see new 750Gb hard drives costing $65 or so (Fry's). $105 for 30Gb is competitive? "Only" forty times more expensive per GB?

If this $20 flash-cache approximates SSD speed on a hard-drive PC, it seems a more economic solution in the short term.


RE: Bottleneck
By Byte on 9/4/2009 8:09:33 PM , Rating: 1
YES! Another layer of caching. So now we have L2, L3, L4 Turbo, Main system memory, RAID controller memory, SSD/HDD DRAM, and then finally to the SSD/HDD. I saw we need 3 more.


RE: Bottleneck
By blakehew on 9/4/2009 11:55:44 PM , Rating: 4
Why is the industry so focused on the SATA 2.5 and 1.8" form factor. Screw SATA lets use the bandwidth of our PCI Express interfaces for SSDs. Lets get more competition there. Right now i think there are just two OCZ and Fuision-IO. I think its dumb that the SSD market is trying to look like the hard drive market. SSDs dont have to look like hard drives. The Hard drive form factor SSDs are nice for laptops, but for my desktop ive got free pci-express lanes let me use them up.


What?
By therealnickdanger on 9/4/2009 10:28:32 AM , Rating: 5
I don't see how this could possibly halt or slow SSD momentum. If implemented properly AND supported by the OS properly, it would offer a performance benefit to systems with slow hard drives and minimal RAM, but by the time this is mainstream, SSDs will be pushing the boundaries of SATA-6Gbps and be cheaper too. With most budget computers coming with at least 3GB of RAM, I can't see where this would even be beneficial. In fact, I would wager that such a system might even hinder the performance of future SSDs.

Unless Intel is completely revamping their implementation of TurboMemory, this will go just as underused as the first rendition.




RE: What?
By HotFoot on 9/4/2009 10:47:07 AM , Rating: 2
I have to agree with you. A buffer is fine for writes, and reading commonly-accessed files, but I fail to see how this technology is going to reduce the time from start-up to being-functional and reduce times for loading programs and all sorts of other tasks.

Last round it was shown that everyone was better off putting the extra money into more RAM instead of turbo memory. By the time this product comes to market, what's the projected SSD cost? We'll probably all be saying everyone would be better off putting the cost of this offering into upgrading from HDD to SSD.


RE: What?
By TomZ on 9/4/2009 10:52:50 AM , Rating: 1
You guys lack any kind of vision. The idea here is to have flash, which is non-volatile, plus have 4X the amount of RAM. If the software was properly written and integrated into the OS, it could probably deliver some nice performance gains and power savings.


RE: What?
By leexgx on 9/4/2009 12:11:04 PM , Rating: 2
but the boot up is mostly I/O intensive, and SSDs do that very well any way real world you would not see an improvement SSD + readybost flash, this type of flash only be use full if your still using HDDs, but should be cheaper then an SSD thats where the benefit is


RE: What?
By leexgx on 9/4/2009 12:13:00 PM , Rating: 2
Turbo Memory not readyboost


RE: What?
By Starcub on 9/4/09, Rating: 0
RE: What?
By therealnickdanger on 9/4/2009 2:56:45 PM , Rating: 2
^
Spoken like someone who has never used an SSD.

I don't mean that in a bad way, just that if you had one you would KNOW how ridiculous that sounds. Yes, SSDs are new and expensive, but not so far out of reach that the average consumer wouldn't be interested. And you certainly do NOT need to have "huge amounts of data (enterprise/professional use)" to realize the benefit of having one. Install one and use it for a day and you'll have a lot of "wow moments". Go back to a HDD system and you'll cry. Seriously.

I, for one, will never go back.


RE: What?
By Jackattak on 9/4/2009 4:31:37 PM , Rating: 2
And for that matter, "huge amounts of data" won't fit on consumer-side SSDs. :) Not at the current capacities, anyway.


RE: What?
By Starcub on 9/7/2009 9:16:32 AM , Rating: 2
quote:
And for that matter, "huge amounts of data" won't fit on consumer-side SSDs.

Nor does your average consumer use RAID 5+ array's, the average consumer can't justify the increased cost given what is available for much cheaper given what is already integrated in modern systems (built in RAID + large capacity HDD's) in default configurations. Thus why I said: SSD is not a consumer tech.


RE: What?
By MrPoletski on 9/7/2009 9:20:51 AM , Rating: 2
The reliability of SSD's will alow you to build big RAID 0 arrays without as much fear of data loss.

RAID 0 is quick, SSD's are quick.

SSD's in RAID 0 are ridiculous.


RE: What?
By leexgx on 9/8/2009 7:42:22 PM , Rating: 2
RAID and SSD is pointless, unlress you Really need the Data rate and IOPs that go with it (Video editing)

in real world use for Home users you not see any improvement as the IOP of an SSD is fast any way and most home users do not run an server from there Home system, you also lose TRIM support with windows 7 as well if RAID is used as Drivers do not support TRIM command yet

1 SSD (with cache ) is faster then an RAID 0 2-3x 15k RPM disk setup


RE: What?
By Starcub on 9/7/2009 9:08:49 AM , Rating: 2
quote:
Spoken like someone who has never used an SSD.

Indeed, nor do I plan to in the near future. SSD's are way too expensive. I can tell just by looking at the numbers that SSD's are several times as fast, at reads, than HDD's. However, the point of my post was that SSD's are not for the general consumer. If they were, they would be reasonably priced, and they aren't. SSD's are still, several years after their introduction to the market, not priced compareably with their performance relative to HDD's. Given SSD marketshare, you can see that the general consumer agrees with me.


RE: What?
By MrPoletski on 9/7/2009 9:18:47 AM , Rating: 2
The only thing you need to fully understand the benefits of SSD drives is to spend two nanoseconds navigating around windows and browsing on an SSD equipped system.


RE: What?
By Yawgm0th on 9/4/2009 2:51:16 PM , Rating: 2
quote:
it could probably deliver some nice performance gains and power savings

It will, but not where it matters. The hard drive is still a bottleneck for a lot of "typical usage" scenarios. If you're going through tons of tiny files (audio, word documents, etc.), the hard drive latency is the bottleneck. A cache will in fact slightly increase the total system latency in those cases, not lower it.

It could be great for some writes, and good for throughput and random access of regularly used files, but ultimately being able to quickly access lots of different files in different places won't be helped. That's where SSDs come in. This is mostly a stopgap.

What we need are affordable consumer-grade hard drives with spindle rates greater than 15,000 RPMs. The premium on high-RPM drives is so great that flash drives will overtake them in capacity over cost ($/GB) at this rate. Why can't I at least get a 250GB 10,000RPM for under $100? Have we really hit such a physical limitation of magnetic hard drive technology that we can't keep increasing spindle rates while lowering costs?

It seems like it. SSDs will win out in a few years, and this technology will die out.


RE: What?
By MRFS on 9/4/2009 3:42:05 PM , Rating: 1
> Have we really hit such a physical limitation of magnetic hard drive technology that we can't keep increasing spindle rates while lowering costs?

No: take a look at the specs for the
Seagate Savvio 15K.2 SAS 2.0 6.0Gbps:

http://www.seagate.com/www/en-us/products/servers/...

http://www.seagate.com/docs/pdf/datasheet/disc/ds_...

Note: sustained transfer rate, outer to inner diameter,
is from 160 MB/second (outer) to 122 MB/second (inner).

MRFS


RE: What?
By MrPoletski on 9/7/2009 9:32:39 AM , Rating: 2
yup and that drive costs nearly $400 at 147GB.

OCZ solid series 120GB SSD costs nearly $300.

160GB Intel X25 goes for around 600-700$ going by google products..

well you decide...


RE: What?
By FaaR on 9/7/2009 12:35:54 PM , Rating: 2
Spinning disk media are in fact largely maxed-out. Increasing spindle speed is not trivial, and increasing head-arm actuator velocity is if possible even less so.

Why else do you think there's hardly any other 15k RPM drives other than Seagate's units still being actively developed, and 15k has been the absolute ceiling for half a decade if not more now.

Even the Raptor series at "merely" 10k RPM only sees occasional updates, and they're priced head and shoulders above your average 7.2k rpm drive.

There's a reason for all this, you know. ;)


RE: What?
By nurd on 9/4/2009 6:15:32 PM , Rating: 2
quote:
Why can't I at least get a 250GB 10,000RPM for under $100?


There's a reason the high-performance drives have lower capacity than the bulk storage drives. And that reason is because they use much lower density platters.

The top size recent drives are using 500 gig platters; 4 platters per 2tb drive. In contrast, WD Velociraptors are using 150 gig platters; 2 for 300 gig drives. Platter sizes in 3.5" 10k SCSI/SAS drives are comparable.

And the reason for that is that what you really want out of performance drives is IOPS; the mass market may be just figuring this out with SSD's, but the high-performance commercial market has known it for decades; this has always been what SCSI drives have pushed for, and why they've always cost so much more.

The less dense your platters are, the wider the tracks are. The wider the tracks, the easier they are to hit and the quicker they are to settle into (and the easier they are to stay in; heads wobble). There are other advantages to the lower density platters, and certainly plenty of other things that add to the cost of the drives, but that's the big reason we have bulk storage drives with 6x the capacity at the same price compared to performance-oriented stuff.


RE: What?
By tygrus on 9/7/2009 2:37:20 AM , Rating: 2
The higher rpm drives although 3.5" on the outside still use smaller (3" to 2.5") platters. This reduces the noise, vibrations, motor size, power requirements, seek times and capacity.

If you are limited to ###MB/s for the heads then MB per track is dependant on the rotational speed:
eg. 120MB/s
5400rpm = 1.33MB per track
7200rpm = 1 MB per track
10K rpm = .72MB per track
15K rpm = .48MB per track

The more tightly packed the data the harder it will be to stay on track. To decrease track-to-track seek times they actually pack less tracks in the same area so the heads have a larger sweet spot and requires less time to get on track. The increase reliability ratings for enterprise drives takes this into account.

There isn't enough sales in the high rpm drives to keep on the bleeding edge of density. They prefer higher reliability and well proven technology.

The reduction in tracks per disk and MB per track reduce the capacity of higher rpm drives.


RE: What?
By kroker on 9/5/2009 2:32:09 AM , Rating: 2
That is exactly what I though when I first read the news, but then I thought about it... If the flash cache stores the most commonly used files, then program loading would seem much quicker for most programs. All of the OS's loading files would eventually be stored because they are accessed often, so this means faster OS boot-up time. Also, all the most used files for the most used applications would eventually get stored in the cache. In the worst case, the system reads from the HDD as usual, so nothing is lost. In the usual case, a few gigabytes (by the way, how large is this cache supposed to be?) worth of most commonly accessed files can be read much faster - which will happen often, because the files are the most used. So, *often* it will be faster than just a HDD.

The advantage vs more memory is that the data stays there after the system restarts. The advantage vs SSD is that you store only the most used files in NAND memory, so you can buy a much larger HDD and it would feel almost like an SSD over time. Even if the experience is not as smooth as an SSD because sometimes the loading is slow and sometimes fast, it will be more often fast than not, depending on your usage pattern. It doesn't have to waste space with large files like movies because they are only watched once or just a few times. Also, unlike an SSD, this is just a cache and not a storage device, so speed is more important than reliability (algorithms like wear leveling may not be as crucial), which may provide some speed optimization options.

And, if you work daily with many different large files, nothing stops you from buying a larger SSD. But I don't think this is the usage pattern of most people.

I think Intel knows what they're doing. Since they'll probably be the only ones to make the flash memory for it, I don't think it's important for them if they sell an SSD or Braidwood, as long as they sell NAND memory in some form.

On average, we would have the storage space of a HDD with the average perceived speed of an SSD. Kind of makes me think why this sort of cache is not implemented in the HDD's themselves though...


RE: What?
By HotFoot on 9/5/2009 3:54:04 AM , Rating: 2
Hybrid HDDs with integrated small amounts of flash were sold a couple years back. I don't think they performed very well at all.

Ideally, for my desktop I'll be sticking with an SSD for the OS and programs, and then media and other personal or work files will be stored on the HDD.

Actually, what I'd like to see is a file sytem that'll take in two or more drives - sort of like RAID, but one drive is SSD and the other HDD. Then I want the system to see these as one drive, but automatically place files over a certain specified size on the HDD and files under that size on the SSD.


RE: What?
By BailoutBenny on 9/5/2009 1:59:47 PM , Rating: 2
Sounds like you want ZFS


RE: What?
By emboss on 9/6/2009 6:05:42 AM , Rating: 2
Yup. IMO flash-based SATA SSDs are the wrong solution to the problem of speeding up storage. By hiding the large-block nature of flash behind a SATA interface, you deny the OS the ability to use the space as effectively as it could.

Turbo Memory and the like are a better solution, if the OS support is there. The OS has much more knowledge about the nature of the data being stored, so can make better decisions regarding wear levelling and the like. There's a heap of files in the typical Windows installation that aren't accessed at all on a particular machine (or are accessed very rarely) so can easily be shunted off to magnetic storage without any problem. On a SATA flash SSD they sit around taking up space (and thus decreasing performance).

Of course, there are a couple of issues. First is the performance. It's going to have to be competitive with SATA drives here, and if it's only got two channels and one chip/channel, that'll be tough. If there's also chips on the back side it could actually be faster than SATA, assuming there's enough planes or stacked die in the package.

The second is it either will increase shutdown times (as potentially gigabytes of modified data gets flushed to magnetic storage) or result in you being unable to move drives between machines (as some of the modified data still sits in the flash).

Finally, I hope Intel is more open about this new implementation than they were with the original Turbo Memory. Intel haven't published a scrap of information as to the hardware interface for that.

Summary: hierarchical storage FTW! :)


How are they going to access this?
By Amiga500 on 9/4/2009 10:31:09 AM , Rating: 2
As above, how are they going to route this to the rest of the mobo?

- Through their direct interconnect?
- Through SATA 3.0?
- Through a dedicated PCI-E pipe?

I guess they know what they are at, and will not starve the bandwidth by making an unsuitable choice.

Anyway, enough of that. For certain engineering problems (I'm thinking Finite Element Analysis here), this could prove to be an inspired move by Intel, as the hard drive is currently the bottle neck on any reasonably powerful workstation (for the likes of FEA... things like CFD are totally different).




RE: How are they going to access this?
By Slayeristight on 9/4/2009 10:41:52 AM , Rating: 2
It will help all hard drives in writing because it is a huge cash but for seeking and loading information off the hard drive you will still be at the same speed of whatever drive you have. So I dont see this having any effect on the sale of SSD drives seeing how they are the fastest thing out there.


By MRFS on 9/4/2009 11:00:36 AM , Rating: 2
http://news.cnet.com/8301-13512_3-10327453-23.html

Braidwood -- like its predecessor, Intel's Turbo Memory technology (formerly code-named Robson) -- is basically a solid-state cache for all the disks in the system .

[end quote]

MRFS


By bhieb on 9/4/2009 11:39:56 AM , Rating: 2
It could help if it had some prefetch algorithms on the read side. Personally it is a nice touch, but IMHO the true glory of the SSD is reading not writing. Sure writing is important, but it is not what most users do most.


By MRFS on 9/4/2009 10:43:29 AM , Rating: 1
That looks like a standard PCI-Express edge connector.

MRFS


Great!!!
By SpaceRanger on 9/4/2009 10:57:31 AM , Rating: 2
I hope these devices aren't gonna be required by the motherboard. If these are NAND devices and are going to see a lot of use, life-cycle of the chips better be REAL good..




RE: Great!!!
By MRFS on 9/4/2009 11:27:53 AM , Rating: 2
Can you say "planned obsolescence" (again)?

MRFS


RE: Great!!!
By SpaceRanger on 9/4/2009 11:32:01 AM , Rating: 2
Just because I can say it doesn't mean I want to see it.. :)


RE: Great!!!
By MRFS on 9/4/2009 11:40:30 AM , Rating: 1
And, will these modules have TRIM support
with sophisticated wear-leveling algorithms?

I'll stick with RamDisk Plus from SuperSpeed:
the DRAM we buy has a lifetime warranty and
in triple-channel mode it moves 25GB/second
at stock settings (no DRAM overclocking).

MRFS


Skeptical
By philosofool on 9/4/2009 11:42:23 AM , Rating: 2
SSD and Braidwood are not competitors. The idea behind a SSD is that you have the contents of your entire HDD on a device with 1/50 the latency of those old platter devices, seek times measured in microseconds rather than milliseconds, and random access times that just blow a HDD out of the water. (Due to space limitations, smart users keep there media files on a traditional HDD. Media files don't need rapid access--do you ever click a song in iTunes and get annoyed by the lag?)

Braidwood looks like a half-way measure at best. The partition on which I keep Vista currently has 55GB space used and I keep all my user data on a different partition. If you want SSD like performance, you're going to need more than 16GB.

Also, MLC an SLC offer the same overall level of performance: the big difference involves the life span of the drive. MLC can last about 1/10th of the time of SLC, which makes a big difference for enterprise class hardware. But a 120GB MLC SSD has a life span of like 500 years of typical desktop use. If it were 4GB, it would have a life of maybe 16 years (1/30 of 120). And if it were being used as a swap file for a desktop machine, it would not be used in a fashion akin to typical desktop use--so you need to make this device with SLC flash to get a good lifespan out to them.

Besides, that CNET article was written by Brooks Crothers, so it probably wasn't even explained correctly.




RE: Skeptical
By philosofool on 9/4/2009 11:47:48 AM , Rating: 2
Oops. I saw the wrong CNET article--there was one a couple months ago by Crothers. Everything else I said was right. (And Crothers is typically confused, but that's irrelevant.)


RE: Skeptical
By Oregonian2 on 9/4/2009 4:18:14 PM , Rating: 1
quote:
do you ever click a song in iTunes and get annoyed by the lag?)


I wonder how much of that is iTunes. When using my 8-year old 1Ghz ancient laptop (has a still nice 1600 x 1200 display), when I drop a MP3 folder onto coolplayer, it plays the first tune pretty much instantaneously, and this is with a slug-slow ancient laptop hard drive. I used to use iTunes before my video-iPod "disappeared", and I found it to be slow at most everything. I was more annoyed at iTunes than the computer's drive system. :-)


RE: Skeptical
By deegee on 9/8/2009 6:33:46 PM , Rating: 2
"But a 120GB MLC SSD has a life span of like 500 years of typical desktop use."

I recommend re-calculating that again...
With cell life, write amplification, wear-levelling, the transient "free space" size, etc., the life span of a 120GB is considerably less than 500 years.
Depending on the drive's specifications and user-use, you're more likely to get something such as 5 years writing 15-20GB per day.

Back to Braidwood though, with small storage sizes such as 4GB, who wants to replace the cache board every 12 months? Not me. Seems a waste and simply a way for mobo manufacturers to get into the "consumables" market like the inkjet printer companies do with the ink cartridges -- now we can force everyone to have to replace their memory twice a year and make a killing in continual after-market sales...


Intel is right
By amanojaku on 9/4/09, Rating: 0
RE: Intel is right
By Spivonious on 9/4/2009 10:30:41 AM , Rating: 5
When considering an SSD for my next system I never once thought of reliability or power consumption. It was all about performance and if the gain was worth the higher price. In the end I decided that it wasn't.


Cache
By SavagePotato on 9/4/2009 10:38:22 AM , Rating: 1
In the end it's still a cache, I don't see how this is better than having a boatload of ram. which many people do nowadays.

If you are starting and application or loading a game level how does this help? It still has to come from your slow magnetic disk on read if it is not already cached.

Granted the concept of superfetch would conceivably keep often used programs in this cache but someone who opens many different applications or games I would sill expect to see the big slowdowns as stuff that is not in the cache is loaded for the first time.

I suppose it would increase OS boot times and stuff like that a great deal but I would rather have the versatility of an SSD that can be upgraded as well in the future as they get faster and larger to some flash soldered on my motherboard. As well as the knowledge that all my applications will launch quickly no matter whether it's the first time I have opened it or the 100th time.




RE: Cache
By leexgx on 9/4/2009 12:19:30 PM , Rating: 2
Turbo Memory under windows vista and 7 use the flash Permanent storage think of the crappy USB readyboost that has to be filled every time the system is booted up but Turbo Memory is different what happens is it not erased every time the system is booted up and windows places files in there so the system uses the Turbo Memory to read the files not the hdd as Braidwood/Turbo Memory is far faster then an hdd more like SSD but it can take 1-2 weeks for windows lern and put stuff on it

if you got SSD (with cache), Turbo Memory is pointless as it most likey be faster to read it from the SSD then the Turbo Memory (windows 7 i guess would turn it off as its based on readyboost tech and windows 7 turns that off when SSD is boot drive)


RE: Cache
By Starcub on 9/4/2009 12:49:52 PM , Rating: 2
quote:
In the end it's still a cache, I don't see how this is better than having a boatload of ram. which many people do nowadays.

I think the cache memory is going to several times larger than the average person has RAM. So both OS and applications will be able to completely load into the cache. The OS would load files necessary for booting, and apps could be precached as they were in Robson. The difference as I see it this time, is that they have increased the space and possibly the bandwidth as well since they are using a different memory tech. IIRC, the primary complaint Robson users had was that the applications that were being cached were not the same as the ones the users were running, thus v2 which allowed for user specified file caching. Perhaps more space, and a better prediction algorithm will make the new tech more user friendly.


great idea needs support
By tastyratz on 9/4/2009 10:45:34 AM , Rating: 2
This is a chicken/egg product with a lot of potential.
Generally I/O intensive operations aren't always with large files so 16gb should do well.
I think its something that probably would benefit more from support in the program itself than a driver. An app can choose to dump to the ssd vs the main drive. The driver can choose to backup information to the main drive after it hits the ssd and detects an idle state for safety.
smaller boot OS files can be loaded on this which are read in parallel with the large files on the main drive for an almost instant boot scenario.

Dram sounds great in theory but the added cost of a battery and complexity probably reserves it to just higher quality HD controllers vs consumer boards.




RE: great idea needs support
By void5 on 9/4/2009 1:17:34 PM , Rating: 2
"Instant boot"? Alas, wishful thinking. Windows (especially Vista and Windows 7) uses a lot of CPU time during boot. Drivers (especially display ones) are also not exactly fast. And you do not really need a SSD to accelerate boot - try beta-version of Boot Cooler (http://forums.anandtech.com/messageview.aspx?catid... ). It can't reach RAM-disk/SSD performance [yet] but is quite close. And I am thinking about adding application launch acceleration ;-)


Why not configure a larger cache in the OS?
By MRFS on 9/4/2009 11:22:12 AM , Rating: 2
What I don't understand, if my reading is correct,
is the role of the existing file cache(s) in
available MS operating systems.

Under "Performance Settings" in XP we have
a choice between "Programs" and "System cache", but
I don't see any way to re-size the latter.

With cheap DRAM and 64-bit operating systems
readily available, why hasn't MS made the
existing file cache more user-configurable?

Properly implemented, those "pages" can also
be "paged" to the swap file "pagefile.sys":
that is how this feature was implemented on
super minicomputers like the DEC VAX 11/780
circa 1980.

Unused DRAM will beat NAND flash every time
especially if the latter is wired to the
PCI-Express bus, but DRAM is installed
in triple-channel DIMM slots with a
raw bandwidth of 10GB+ per second.

MRFS




By MRFS on 9/4/2009 12:36:51 PM , Rating: 2
Let me illustrate with a task that we perform
quite frequently: XCOPY folder X:\folder /s/e/v/d/l

"X:" is an XP network drive wired to Gigabit Ethernet
LAN adapters and a Linksys Gigabit switch.

When we do this same task twice, first after a cold start,
then second immediately after the first one finishes,
we always see a much higher data rate across the GbE LAN:

http://www.supremelaw.org/systems/io.tests/first.a...

The differences between the first and second runs are
even more pronounced when the source and target disks are
both NON-RAID. In the graph above, both source and
target folders were hosted on RAID-0 partitions.

My theory to explain these consistent differences
is this: the "/l" command line option does not require
XCOPY to read anything but folder entries, because
XCOPY only compares file sizes and dates last modified.

XOPY does not do a bit-wise comparison of each
corresponding file pair (source and target).

In the second of 2 runs above, those folder entries
are memory-resident -- in XP's system I/O cache --
whereas in the first of those 2 runs those folder entries
must be read from the hard drives at source and target.

MRFS


By RMSe17 on 9/4/2009 10:50:31 AM , Rating: 3
Sounds like the guy has investments with SSD manufacturers :)




By dwmccauley on 9/4/2009 10:52:42 AM , Rating: 2
What about the report in DigiTimes and elsewhere that Braidwood & P57 were cancelled?
"Intel said to not include Braidwood in upcoming 5-series chipsets"
http://www.digitimes.com/NewsShow/MailHome.asp?dat...




By nemitech on 9/4/2009 11:09:27 AM , Rating: 2
It woudl be better if this technology was on a PCIe controller card with e.g. 4 SATA (RAID) ports behind it. That way you could upgrade old / existing systems and easily migrate to new ones. I wonder how long until (raid) controller manufactures start to implement simialr cache on their offerings.




By maddoctor on 9/4/2009 12:15:25 PM , Rating: 2
Intel as the only innovator on computing market have been planned the future technology that could generates money for them and their owners. Intel is always true with SSD market position and will owned the majority marketshare of SSD. This opportunity will be able to drive the computing powers potential without bottleneck. Combination of Micron and Intel in SSD development will bring some competitors to running out of cash. I believe Intel monopolies in this market will benefits the consumer especially that bought Intel products. I believe everything is Intel Inside.




By Mjello on 9/4/2009 12:38:09 PM , Rating: 2
Funny how seceral sites disagree with daylytech

http://www.pcstats.com/NewsView.cfm?NewsID=78087

http://www.digitimes.com/news/a20090819PD217.html

I dont know what to believe, but this article seems to relate to old news from june 2.

Apparently intel has canceled or prosponed the product.




By XZerg on 9/4/2009 12:38:33 PM , Rating: 2
quote:
Capacities will be approximately 4GB to 16GB, and the cost increase will be approximately $10 to $20 per system, according to analyst Jim Handy, who authored a recent report on Braidwood.


Let's for argument sake this increases $20 for 16GB then what stops motherboard makes to charge extra $40 and have 2 of 16GB on the motherboard? That's a huge performance jump for very little cost comparatively.

But then again this all comes down to the quality of the controller. Keeping in mind that most SSD drive makers are using multi-nand chips to improve performance which 1 16GB chip would have tough time matching up to.

Let's wait and see how this progresses... next year definitely seems like the year for a lot of stuff.




Show me the number...
By Jeff7181 on 9/4/2009 3:15:18 PM , Rating: 2
I can see the benefit as a write cache, but as a read cache I don't believe it will be as effective. It would depend largely on some type of algorithm that can accurately predict what data you need next. Otherwise if it's not in the cache, you're still waiting for heads to move and platters to spin.

Don't get me wrong, I've supported this type of technology for a while now and have been confused why hard drive caches are so small (comparative to the HDD's capacity). The reason I keep finding for <32 MB caches on hard drives is that you reach a point of diminishing returns. By adding more cache and more processing power to process the algorithms to use the cache effectively, it becomes cheaper to just add an additional spindle.

Definitely will not reduce demand for SSD's.




Good idea, wrong place
By gstrickler on 9/4/2009 4:03:40 PM , Rating: 2
16GB of SLC flash as a semi-permanent cache for a HD could be great, as long as it uses a good caching algorithm. That's plenty to cache the most used parts of the OS, file system, and applications, as well as commonly accessed data/configuration files. Once it's tuned to your usage patterns, you could conceivably boot, do some work, and shutdown without ever reading from the HD.

The problem with "Turbo Memory" and with "Braidwood" is that they require OS level drivers/support, figuring out which data to put on the cache can be tricky, and it can change over time as usage patterns change. Second problem is that when you update data, you have to update the cache and the device it's it's providing caching for.

Move this to the disk channel itself by including it in the SATA/RAID controller or on the drive itself, and you can make it OS independent. If it's on the drive, the caching algorithm can be tailored to address the physical properties of the drive (HD, optical, SSD, etc), and you can eliminate the double write (or as least defer it to idle time). If HDs continue to be significantly higher capacity and/or lower cost/GB than SSDs, this could be useful.




By corduroygt on 9/4/2009 5:24:07 PM , Rating: 2
Along with a $100 1.5tb 3.5" or 500gb 2.5" drive, they can also offer $125 versions of the same drives with 16GB cache. It'd be a nice mid-range solution before SSD's become truly affordable.




Windows 7 SP1
By vwgtiron on 9/8/2009 1:21:00 AM , Rating: 2
Isn't this technology coming out about the time that a windows 7 SP1 would be to? That would give you the OS support for the new cache technology if Intel pays enough!




By William Gaatjes on 9/8/2009 2:10:16 PM , Rating: 2
From the text :

quote:
Intel's upcoming Turbo Memory successor, Braidwood, will consist of NAND flash module residing on "5 Series" motherboards (used with the upcoming Westmere 32 nm processors) and serve as a cache for all reads and writes .


I hope they also improved the endurance of the flash memory.
Because wear leveling and other techniques to prevent the flash memory cel from degrading fast will not be enough. I know modern operating systems already caches and schedules writes to the HDD to improve performance but still, i doubt that flash technology is improved that much. I see somebody creating a new market...

A controller with 1GB of DDram memory for write caches and flash as read cache together with a traditional HDD for bulk storage would be much better. Design the controller that it can write the data in the DDram to the HDD on it's own if the OS desires it. A sort of display list but for HDD controllers. Maybe it is time that the HDD controller get's a little smarter and more versatile. A HDD controller with it's own DRAM interface and it's own flash memory interface and the traditional but stripped down for pure speed version of a HDD interface. It would be like moving the hdd controller on the HDD back to the motherboard again. Keeping only the necessary logic / ram and analog circuits on the HDD. Away with al the overhead.

Example : The lynnfield has a pci express controller on board. A second one could be used only for the data controller connected to the HDD ,flash and dram.
Or use A hyper transport connection like AMD has.
Strip it down enough to gain a low latency but sufficiënt in speed for the upcoming years.

We would have a system build up of a CPU, GPU and a DPU (data processing unit) :). The DPU and main memory will in the future be fused together and be the same .




"When an individual makes a copy of a song for himself, I suppose we can say he stole a song." -- Sony BMG attorney Jennifer Pariser

DailyTech Poll
Do you use copy/paste on your smartphone? 




14 Comments









botimage
Copyright 2010 DailyTech LLC. - RSS Feed | Advertise | About Us | Ethics | FAQ | Terms, Conditions & Privacy Information | Kristopher Kubicki